SemDiff

Semantic diff for documents.
One API call.

Two document versions in — clause-level changes out, as structured JSON. Not a visual redline: data your pipeline can act on.

Get early access See the API

Founding developers: $49 deposit = 30% off for life, refundable anytime

In, and out

# request
curl -X POST https://api.semdiff.dev/v1/diff \
  -H "Authorization: Bearer $KEY" \
  -F "base=@msa_v1.pdf" -F "revised=@msa_v2.pdf"

# response (truncated)
{
  "summary": { "changes": 14, "high_impact": 3 },
  "changes": [
    {
      "type": "modified",
      "section": "9.2 Limitation of Liability",
      "base":    "...capped at 12 months of fees",
      "revised": "...capped at 3 months of fees",
      "impact": "high"
    },
    {
      "type": "added",
      "section": "14.7 Non-Solicitation",
      "impact": "high"
    }
  ]
}

Built for document pipelines

Clause-aware matching

Sections matched by meaning and structure, not line position. Reordered ≠ changed.

Tables & scans

Cell-level table diff. OCR built in — scanned PDFs welcome.

Deterministic JSON

Stable schema, severity scoring, section anchors. Pipe it to Slack, your CLM, or a review queue.

Why not the alternatives?

	SemDiff	Visual diff tools (Draftable etc.)	DIY with an LLM
Output	Structured JSON	Rendered redline for humans	Unstable, prompt-dependent
Entry price	$29/mo self-serve	~$1,000/yr API minimum	Your eng time
Tables / scans	Cell-level + OCR	Limited	Hard to get right

Pricing

Dev

50 diffs/mo
10 MB files
Community support

Starter

$29/mo

500 diffs/mo
50 MB files
Email support

Pro

$99/mo

3,000 diffs/mo
Priority queue
Webhooks

Early access

Founding deposit — $49 (30% off for life)

Fully refundable until you make your first API call.

FAQ

When does the API launch?: Private beta is rolling out to founding developers first. Deposit holders get access in cohort order.
What formats are supported?: PDF (digital + scanned), DOCX at launch. XLSX and HTML on the roadmap.
Do you store my documents?: Processed in memory where possible; any temporary storage is encrypted and purged within 24 hours. Never used for training.
How is this different from text diff?: Text diff compares characters. SemDiff aligns sections, clauses, and table cells by meaning — a renumbered section isn't 40 false changes, and a quietly reworded liability cap isn't buried in noise.

Semantic diff for documents.One API call.